Saturday, December 14, 2013

A Puzzle For You: What Drives The Findings In the Race/Ethnicity Internet Dating Study?


The point of talking about this:  Everywhere I checked people assumed that the data correctly reflects not only race or ethnic differences in how people respond to interest expressed by others but also that it reflects actual racial or ethnic preferences of the type where the choices shown in the picture below are how people would choose a partner in real life.

I saw only a few points of criticism about the study.  Most of us simply accept that it is correctly done.  This, in my opinion, puts a higher responsibility on the shoulders of those who pass these studies into the popular media.  Specifically, the actual study should be easily available and proper discussions about it should be encouraged, not just cheap clickbaits.

 -----


I missed this study (?)* which uses data from the Facebook dating app to draw conclusions about racial or ethnic preferences in heterosexual dating.  Here's a recent summary of its findings:

Quartz, a business and marketing website, recently released data on the Facebook dating app Are You Interested, which connects single people with others within the confines of their Facebook networks. Quartz’ data are based on a series of yes-or-no questions about who users are interested in, as well as response rates between users, once notified of a potential suitor. The data show that white men and Asian women receive the most interest, whereas black men and women receive the least amount of interest. The writers at Quartz summarize the findings as follows:
Unfortunately the data reveal winners and losers. All men except Asians preferred Asian women, while all except black women preferred white men. And both black men and black women got the lowest response rates for their respective genders.
I was still confused about what the study used as its measure.  A better description for how the data was created is here, attached to a picture which purports to show the highest and lowest response rates.  Here's the picture (click to make larger):


And here's the explanation:

The data shown above come from the Facebook dating app, Are You Interested (AYI), which works like this: Users in search of someone for a date or for sex flip through profiles of other users and, for each one, click either “yes” (I like what I see) or “skip” (show me the next profile). When the answer is “yes,” the other user is notified and has the opportunity to respond. It’s very similar to another dating app, Tinder.

The graphic shows what percentage of people responded to a “yes,” based on the gender and ethnicity of both parties (the data are only for opposite-sex pairs of people). Unsurprisingly, most “yes’s” go unanswered, but there are patterns: For example, Asian women responded to white men who “yessed” them 7.8% of the time, more often than they responded to any other race. On the other hand, white men responded to black women 8.5% of the time—less often than for white, Latino, or Asian women. In general, men responded to women about three times as often as women responded to men.
 I have emailed the site to find out if I can have the actual study so as to find its methodology etc.  Couldn't find it anywhere on the net.

Something smells off in all this**.  Perhaps that's because most people in real life seem to partner with people who share the same racial or ethnic background?  Then there is the two-rounds aspect of the dating game:  First people have to click "yes" on your profile.  Second, you decide if you wish to respond to that "yes" or not.

But the first round (where the first contact is made through that "yes") probably also has  differences by racial or ethnic group membership, and I haven't seen anything about how that part is taken into account.  If it is.  And the various groups that picture shows are most likely not present in equal numbers in the Facebook dating app.  I have tried to play around with hypothetical group sizes and hypothetical racial and ethnic preferences and different guesses about what percentage of "yeses" people might respond to, in general, and with respect to their own group vs. other groups.***

I'm not getting very far with that.  Or rather, I can get almost anywhere because I don't have the data I need to have.  And that the only data easily available is about those extremes also troubles me.  I'd like to see the average values, sizes of the groups and so on, and I would dearly want to see the same analysis about any racial or ethnic difference in the first round assignment of "yeses."
------
*The question mark is because I'm not sure that this is a study, in the academic sense of the word, with a research paper which explains how the data was obtained, what standardizations were carried out etc..
**Not necessarily "wrong," but all this looks far too simplified (and the picture picking the "least" and the "most" preferred connections ignores all the other data).  If there's racial or ethnic preference in the first round when "yeses" are given, then that is completely ignored by focusing only on the second round (responding to the "yeses").  I'm also not sure if "preference" is the right word to use in this context.
***For instance, suppose that the largest number of "yeses" for some profile comes from people who share that profile owner's race or ethnicity.  Say that group is a hundred "yeses."  Suppose that this profile also gets five "yeses" from people of a different race or ethnicity.  Suppose the owner then responds to, say, twenty of the "yeses" coming from his or her own group but to all the five "yeses" coming from the other groups.  Then the raw results would suggest that this person has a 100% response rate for people outside his or her race or ethnic group but only a 20% response within the same racial or ethnic group.

Which would be correct but would not really mean that this person has a preference for dating people from other racial or ethnic groups.  So I think we need to know the numbers in much greater detail than they are given in the summaries I've been able to read.